With the release of Intel’s Meteor Lake CPUs to the market, data flooded in from all quarters, and we previously warned that many reviews would need to be redone due to numerous inconsistencies. This is what happens with a new, disruptive platform: software has to adapt to hardware, and that takes time. Well, with some BIOS changes and Windows patches, performance improved, but on Linux… It was assumed that these issues weren’t present, but Intel has shown that wasn’t the case. As a result, Meteor Lake CPUs will see a 100 MHz increase on Linux 6.8, so what about the previous comparative data?
Indeed, as older people say, to heck with it. However, to be specific, not all data will undergo an update per se. Interestingly, the problem identified by Intel has not been detected by any analysts until now, and it is, of course, related to the OS’s power states.
Linux 6.8 will see Intel Meteor Lake CPUs scale 100 MHz
Interestingly, the issue does not occur on all systems running Linux, or at least that’s what Intel specifies. To understand the message from the chipmaker, let’s briefly summarize the problem.
It seems that everything revolves around Linux’s power state and the so-called Intel P-State CPUFreq driver, which is unable to achieve the CPU’s full frequency due to a simple scaling error. Intel’s Rafael Wysocki explains it as follows:
“On some Meteor Lake platforms, the maximum single-core turbo frequency is not observed. During hybrid performance-to-frequency conversion, this is 100 MHz lower than it should be. This results in a maximum requested frequency 100 MHz lower.
For example, when the maximum single-core turbo is 4.9 GHz: MSR HWP_CAPABILITIES shows that the highest performance ratio for the P-core is 0x3E. With the current scaling factor of 78741 (1.27x to convert frequency to performance), a maximum frequency of 4.8 GHz is obtained. This results in a maximum scaling frequency limit of 4.8 GHz, which is 100 MHz less than desired.
Add the ability to define a CPU model-specific scaling factor and set the scaling factor to 80000 (1.25x to convert frequency to performance for P-cores) for Meteor Lake. This will restore the frequency and performance.”
How much will this change the performance of these 3D-manufactured CPUs?
In principle, not much compared to what we saw earlier, mainly because the improvement is understood to be directed at single-thread rather than multi-thread performance. The issue is that Wysocki’s statement and explanation leave the door open for this 100 MHz increase to apply to all P-cores, not just one, due to the problem with Linux’s power manager and its controller.
Regardless, corrected data will arrive next week with the patch that will boost Meteor Lake’s frequency by 100 MHz on Linux 6.8, and we’ll be able to clear up any doubts. This is interesting, as in the previous comparison, AMD’s Ryzen 7 7840U won almost 80% of the tests, which is quite a significant balance compared to what was observed on Windows.
Therefore, this patch could bring the percentages closer together and show more balanced performance on Linux 6.8; we just have to wait a week.
The post Intel Meteor Lake CPUs to be faster on Linux 6.8: An OS error lowered them by 100 MHz first appeared on El Chapuzas Informático.